It was during my time in college when the first discrete graphics cards came out. The fast and improved graphics of Quake
sold me on picking up a Voodoo card. Before then, the CPU did all the work and the graphics cards had specific operations to handle graphics. These days, the cards that AMD and NVIDIA are putting out have become more generic in their nature being able to handle different tasks. Even physics acceleration has been shown to work on these cards. So, in my opinion, graphics cards are becoming a secondary CPU being able to handle complex calculations and whatever job they are asked to do. No longer does one part just deal with textures and another part deals with verticies.
So it doesn't surprise me when Intel is talking about their Larrabee chip
aimed at taking on the two big graphics card giants. We're talking 8 to 48 even hundreds of cores (down the road) to offer increased graphical performance. It's based on the original Pentium design and not the Core architecture in the current CPUs. Anandtech has some good information
and there's a lot to digest. Intel makes some bold statements such as linear improvement on performance. Not even dual card configurations today offer a true doubling of performance. So for Intel to say a 16 core Larrabee will double the performance of an 8 core Larrabee is pretty big. They also claim when it's released it will have performance comparable to the GPUs at that time.
Now the chip is not due out for another year or so so everthing today and in the near future is speculation. While Intel sells a lot of their integrated graphics units, they certainly aren't the best performers and are never recommended for gaming. Can the guys behind the current leader in CPU performance be a player in the GPU arena as well? We shall see but I think it's going to take a lot to match up with the two current giants in the industry. But if anyone can do it, Intel has the knowledge and resources to do it.